This is the question worth asking before any of the others, because the failure that actually damages a business is not silence. It is a confident wrong answer given to a real customer in your name.
Book a free discovery call and hear what it would sound like on your line.
That is our AI receptionist answering for a firearms training school. Nobody walked it through that answer. It just knows the business.
It knew the course inside out. And the one thing it was not sure about, it did not invent an answer. It got a human involved.
Most people evaluating AI on the phone worry about the wrong failure. They picture it not understanding, going quiet, or annoying somebody. Those are irritating, and they are recoverable.
The failure that does lasting damage is different: a fluent, confident, completely wrong answer, delivered in your business's name to somebody who then acts on it. A course prerequisite that does not exist. A turnaround time you cannot meet. A statement about what is included that you will now have to honour or awkwardly retract.
In regulated or safety-adjacent work, that stops being a service problem. Firearms training, veterinary advice and anything touching compliance all carry answers that must be right, and the cost of a plausible invention is not a lost booking, it is a misinformed customer.
The call above is a real one against a firearms training school's system. The caller's voice has been re-recorded for privacy; the AI's side is unedited.
Two things happen in it, and the second matters more than the first.
Course content, structure, what the training covers. Nobody prompted it through the answer during the call and nobody scripted that specific exchange. It answered from what it had been built to know about the business.
On the one point it was not certain about, it did not produce something plausible. It said it would get the caller a proper answer, took their contact details, and passed the question to a person on the team.
That is the behaviour to look for, and it is a design decision rather than an accident. A system can be built to always produce an answer, or to recognise the boundary of what it reliably knows and hand over there. The first demos better. The second is the one you want answering your phone.
There is an assumption that a handoff is a failure of the system. It is worth questioning, because it is not how anyone judges a human receptionist.
A good human receptionist does exactly this. They handle the routine confidently, and on the unusual question they say they will check and come back to you. Nobody considers that a poor experience. It reads as competent and honest, and the caller gets a correct answer slightly later instead of a wrong one immediately.
The bar for the AI is the same. Handle the great majority of calls end to end, and on the remainder, capture everything needed and route it to someone who knows. What you should not accept is a system that treats every question as one it must answer.
Where the line sits is yours to decide, and it is worth deciding deliberately rather than discovering it later.
Anything with legal, medical, safety or compliance weight belongs on the human side by default. So does anything where being wrong is expensive to unwind, and anything you would not want stated without your judgement applied. Everything else, the routine and repeated questions that make up most of your call volume, the system handles.
The handoff itself should be a real one rather than a dead end. The caller's details, the question, and enough context that whoever picks it up is not starting from nothing. Otherwise you have replaced an invented answer with a lost message, which is a different failure rather than a fixed one.
One honest note on the above. Some of this is included on every plan and some of it, particularly text messaging, bilingual answering, self-service rescheduling, no-show rebooking and the review engine, depends on which plan you are on. We map the pieces you actually need on the call rather than selling you the whole list.
It should stop rather than improvise: tell the caller it will get them a proper answer, capture their details and the question, and route it to a person. In the recording above that is exactly what happened on the one point it was not certain about.
That is the risk worth asking about, and it is a design decision rather than luck. The system is built around what it reliably knows about your business, with anything outside that boundary handed to a human instead of answered. You decide where the boundary sits.
No more than it is when a good human receptionist says they will check and come back to you. The caller gets a correct answer slightly later instead of a confident wrong one immediately, which is the better outcome for both sides.
Anything with legal, safety, medical or compliance weight, anything expensive to unwind if it is wrong, and anything you would not want said without your judgement applied. Routine, repeated questions are what the system is for.
Book a free 30-minute discovery call and push it on the questions that matter in your business.
Book Free Discovery Call